YFarmX logoYFarmX

AI Desk · AI Fingerprinting · Compare

DeepSeek: DeepSeek V4 Pro 0813 · Tencent: Hy4 preview

Identity similarity: lowDivergent fingerprints · 28/50These models tokenise differently. The rows that part here are the English prose, whitespace, CJK and non-Latin script strings. Divergence on high-information strings separates unrelated vocabularies within a handful of samples.

DeepSeek: DeepSeek V4 Pro 0813 recordTencent: Hy4 preview recordOpen in the interactive bench

View all 50 test strings28 matched · 22 different · 0 excluded

String-by-string results

Expand a string to read or copy its exact text. Each row compares the two models’ token counts.

28 of 50 clean strings agree

Showing all 50 strings.

Test stringDeepSeek: DeepSeek V4 Pro 0813Tencent: Hy4 previewResult
English sentenceen-prose
Exact text
The quick brown fox jumps over the lazy dog near the riverbank.
1414✓ Matched
Long English sentenceen-long
Exact text
Attribution of a stealth model requires evidence that the model cannot choose to withhold.
1617≠ Different
20 spacesspaces-20
Exact text
a                    b
Show whitespace marks

· = space, ⇥ = tab, ↵ = line break

a····················b
34≠ Different
60 spacesspaces-60
Exact text
a                                                            b
Show whitespace marks

· = space, ⇥ = tab, ↵ = line break

a····························································b
36≠ Different
20 tabstabs-20
Exact text
a																				b
Show whitespace marks

· = space, ⇥ = tab, ↵ = line break

a⇥⇥⇥⇥⇥⇥⇥⇥⇥⇥⇥⇥⇥⇥⇥⇥⇥⇥⇥⇥b
411≠ Different
20 line breaksnewlines-20
Exact text
a



















b
Show whitespace marks

· = space, ⇥ = tab, ↵ = line break

a↵
↵
↵
↵
↵
↵
↵
↵
↵
↵
↵
↵
↵
↵
↵
↵
↵
↵
↵
↵
b
47≠ Different
Mixed whitespacemixed-ws
Exact text
a 	
  	
   b
Show whitespace marks

· = space, ⇥ = tab, ↵ = line break

a·⇥↵
··⇥↵
···b
77✓ Matched
9 digitsdigits-9
Exact text
123456789
33✓ Matched
12 digitsdigits-12
Exact text
123456789012
44✓ Matched
30 repeated digitsdigits-30
Exact text
111111111111111111111111111111
1010✓ Matched
Formatted numberdigits-sep
Exact text
1,234,567.89
77✓ Matched
Long decimalfloat-long
Exact text
3.14159265358979323846
99✓ Matched
Chinese sentencezh-common
Exact text
这是一个用于测试分词器行为的中文句子
98≠ Different
Long Chinese sentencezh-long
Exact text
人工智能模型的归属需要可验证的证据而不是自我报告
1010✓ Matched
Rare Chinese characterszh-rare
Exact text
龘齉齾爩鱻麤靐飝鞻
1818✓ Matched
Japanese kanaja-kana
Exact text
これはトークナイザーのテストです
910≠ Different
Japanese kanjija-kanji
Exact text
日本語の漢字と仮名の混在文章
1111✓ Matched
Koreanko
Exact text
한국어 토크나이저 테스트 문장입니다
1312≠ Different
Russianru
Exact text
Проверка токенизатора на кириллице
1313✓ Matched
Arabicar
Exact text
اختبار المحلل اللغوي للنصوص العربية
1012≠ Different
Hebrewhe
Exact text
בדיקת מנתח אסימונים בעברית
1414✓ Matched
Hindihi
Exact text
यह टोकनाइज़र परीक्षण वाक्य है
1817≠ Different
Thaith
Exact text
นี่คือประโยคทดสอบตัวแบ่งคำ
1114≠ Different
Greekel
Exact text
Δοκιμή του τμηματοποιητή κειμένου
1620≠ Different
Basic emojiemoji-basic
Exact text
😀😃😄😁😆
1010✓ Matched
Emoji skin tonesemoji-skin
Exact text
👍🏻👍🏼👍🏽👍🏾👍🏿
1414✓ Matched
Joined family emojiemoji-zwj-family
Exact text
👨‍👩‍👧‍👦
1111✓ Matched
Repeated family emojiemoji-zwj-x3
Exact text
👨‍👩‍👧‍👦👨‍👩‍👧‍👦👨‍👩‍👧‍👦
3333✓ Matched
Flag emojiemoji-flags
Exact text
🇬🇧🇺🇸🇨🇳🇯🇵
1616✓ Matched
Profession emojiemoji-prof
Exact text
👩🏽‍💻👨🏿‍🚀🧑‍🔬
1919✓ Matched
Mathematical symbolsmath
Exact text
∀x∈ℝ ∃y∈ℚ : x⊕y ≡ ∅ ∧ ∇²φ = 0
2227≠ Different
Box-drawing charactersboxdraw
Exact text
┌─┬─┐│├┼┤└┴┘╔═╗╚╝
2728≠ Different
Combining accentscombining
Exact text
áêĩōü
1010✓ Matched
Extended Chinese characterscjk-ext-b
Exact text
𠀀𠀁𠀂𪛖
1616✓ Matched
Styled Unicode letterssurrogates
Exact text
𝕳𝖊𝖑𝖑𝖔 𝓦𝓸𝓻𝓵𝓭 𝟙𝟚𝟛
3341≠ Different
Layered combining markszalgo
Exact text
h̸̢̛e̵̛�݃l̷̨͐l̸̛͝o̶̧͝
3636✓ Matched
Mixed writing directionsrtl-mix
Exact text
hello שלום world مرحبا
67≠ Different
Python codepy-code
Exact text
def f(x):
    return [i**2 for i in range(x) if i % 3 == 0]
Show whitespace marks

· = space, ⇥ = tab, ↵ = line break

def·f(x):↵
····return·[i**2·for·i·in·range(x)·if·i·%·3·==·0]
2525✓ Matched
Indented Pythonpy-indent
Exact text
if a:
        if b:
                if c:
                        return 1
Show whitespace marks

· = space, ⇥ = tab, ↵ = line break

if·a:↵
········if·b:↵
················if·c:↵
························return·1
1516≠ Different
JSON objectjson
Exact text
{"model":"ox-alpha","context_length":1048576,"top_p":0.95}
2020✓ Matched
HTML markuphtml
Exact text
<div class="row"><span id="a">x</span></div>
1616✓ Matched
Regular expressionregex
Exact text
^(?:[a-z0-9!#$%&*+/=?^_`{|}~-]+)@(?:[a-z0-9](?:[a-z0-9-]*)?\.)+$
4746≠ Different
Camel-case identifiercamel
Exact text
getUserAccountBalanceByIdentifier
76≠ Different
Snake-case identifiersnake
Exact text
get_user_account_balance_by_identifier
86≠ Different
Repeated long wordrare-word-x5
Exact text
antidisestablishmentarianism antidisestablishmentarianism antidisestablishmentarianism antidisestablishmentarianism antidisestablishmentarianism 
2631≠ Different
Repeated wordrepeat-tok
Exact text
lorem lorem lorem lorem lorem lorem lorem lorem lorem lorem lorem lorem lorem lorem lorem lorem lorem lorem lorem lorem 
4141✓ Matched
Base64 textbase64
Exact text
SGVsbG8gV29ybGQhIFRoaXMgaXMgYSB0ZXN0IHN0cmluZy4=
3435≠ Different
Hexadecimal texthex
Exact text
deadbeefcafebabe0123456789abcdef
1212✓ Matched
URL texturl
Exact text
https://openrouter.ai/api/v1/models?limit=100&offset=0
1717✓ Matched
UUIDuuid
Exact text
3f2504e0-4f89-11d3-9a0c-0305e82c3301
2727✓ Matched

Counts show each string’s prompt-token cost after the fixed overhead is subtracted. Excluded readings are kept separate from matched and different results.

Published model settings

Limits, prices and capabilities

Compare the essentials, then open the developer settings for more detail.

ADeepSeek: DeepSeek V4 Pro 0813
BTencent: Hy4 preview

Context window

Same

Total token capacity per request

A1,048,576 tokens
B1,048,576 tokens

Maximum reply

Different

Published output limit in tokens

AUnspecified
B64,000 tokens

Input price

Different

US dollars per 1 million input tokens

AUS$0.40 / 1M tokens
BUS$0.7506 / 1M tokens

Output price

Different

US dollars per 1 million output tokens

AUS$5.00 / 1M tokens
BUS$2.2509 / 1M tokens

DeepSeek V4 Pro 0813: OpenRouter's listing shows 943,718 output tokens, which is 90% of the context window, the figure it carries where the provider declares no maximum. DeepSeek's own API reference caps max_tokens at 384K, 393,216 tokens, for deepseek-flash and deepseek-v4-pro.

What you can send and receive

Same

ADeepSeek V4 Pro 0813

Accepts
text
Produces
text

BHy4 preview

Accepts
text
Produces
text

Developer settings

Open a section to compare the complete declarations.

Supported API settingsShared controls and the settings unique to each model18 shared · 3 model-specific

AOnly DeepSeek V4 Pro 0813 2

  • Token probabilitieslogprobs
  • Top token probabilitiestop_logprobs

BOnly Hy4 preview 1

  • Maximum completion tokensmax_completion_tokens
18 settings supported by both models
  • Frequency penaltyfrequency_penalty
  • Include reasoninginclude_reasoning
  • Token biaslogit_bias
  • Maximum tokensmax_tokens
  • Minimum probabilitymin_p
  • Presence penaltypresence_penalty
  • Reasoningreasoning
  • Reasoning effortreasoning_effort
  • Repetition penaltyrepetition_penalty
  • Response formatresponse_format
  • Seedseed
  • Stop sequencesstop
  • Structured outputsstructured_outputs
  • Temperaturetemperature
  • Tool choicetool_choice
  • Toolstools
  • Top ktop_k
  • Top ptop_p
Reasoning settingsEffort levels and reasoning behaviour2 different · 2 shared
A3 declared fieldsB4 declared fields
SettingADeepSeek V4 Pro 0813BHy4 preview
Reasoning requiredNoNo
Effort levelsmaxhighlowhighlownone
Default efforthighhigh
Enabled by defaultUnspecifiedYes
Default settingsValues used when a request leaves a setting open2 different · 0 shared
A2 declared fieldsBNo values declared
SettingADeepSeek V4 Pro 0813BHy4 preview
Temperature1Unspecified
Top p1Unspecified

Published API declarations. Prices and limits can vary by provider.

Method: tokenizer fingerprinting, v1, 50 strings. This page is static and dated; the interactive bench compares any two of the 353 measured models.